In general, there are three ways in which a molecule with novel properties may be obtained. A first method, e.g., protein engineering, relies on known properties of a general type of molecule and upon theoretical models which attempt to define the conformation of molecules most likely to have the desired properties. No models have proved general enough or exact enough to reproducibly design appropriate molecules.
A second method is screening. Screening requires that multiple permutations of molecules be tested for a given property. The current status of screening technology and the vast number of different permutations limits the usefulness of this technique. For example, a peptide sequence of twenty amino acids has 20.sup.20 different permutations. To screen bacteria producing different permutations of peptides of significant length, billions upon billions of petri dishes, each on the order of a thousand colonies, would be needed. To screen such large populations to find those few members, if any, which have the desired characteristics is extremely inefficient. Screening techniques are not adequate for the realistic performance of such tasks.
A third method employs natural selection in specific non-generalizable ways. For example, if a unicellular organism is missing an enzyme in a critical metabolic pathway, one can try to select for a molecule with the same function as that lost by the mutant. This technique is limited, however, by the reactions that are encoded in the genome of the organism and that may be complemented within the cell. Moreover, for each different complementation experiment, a new mutant strain is needed.